3 Investment Ideas With Room to Run

AI agent adoption is expected to sharply raise inference demand (Gartner: 5x–30x more tokens vs single-shot; Goldman: up to ~120 quadrillion monthly tokens by 2030), creating a “compute toll-road” investment focus across accelerators/networking/memory/storage. Separately, SK Hynix plans a $28B IPO with demand ~7x available shares, signaling strength for the AI memory trade, while U.S. initial jobless claims fell to 215,000 (below forecasts), indicating a steady labor backdrop. For non-AI exposure, Middle East disruption is driving refinery margins—3-2-1 crack spread up 73% in Q1—and Phillips 66 (PSX) delivered adjusted EPS of $0.49 vs an expected loss of $0.39; the consensus has been revised up 60% with Q2 earnings projected at $6.64 (+179% YoY).

Analysis

Agentic AI is likely to shift spend toward inference-heavy infrastructure, but the market is still underpricing where the margin pool migrates. Near term, the fastest beneficiaries are not necessarily the headline GPU names: CRDO and COHR should see outsized revenue leverage if token volumes compound, while P looks like a higher-quality way to own storage because recurring subscription revenue cushions hardware cyclicality. DELL benefits on unit shipments, but it is the least levered economically because systems assembly is the most competitive layer of the stack.

The bigger second-order risk is that more tokens do not translate one-for-one into revenue if hyperscalers and enterprises optimize models, compress context windows, or shift workloads to cheaper custom silicon. That argues for buying the "picks and shovels" basket on weakness rather than chasing after a sentiment spike; the supply-tightness signal in AI memory is useful, but still a sentiment indicator, not a cash-flow guarantee. Over 6-18 months, the winners should be the vendors sitting on persistent bottlenecks: memory, optical interconnect, and storage.

On energy, the market is likely over-leaning into PSX as a clean geopolitical hedge after a sharp run. Refining margin support can stay elevated for weeks, but the thesis weakens quickly if diplomacy lowers product shortages or if crude spikes enough to hit end-demand; that makes this a 1-3 month trading vehicle, not a structural long. Brazil is the most contrarian angle, but it is a macro commodity-and-power proxy, not an immediate AI monetization story, so I would treat EWZ as a watch item tied to commodity prices and local policy rather than chase it outright.

Key Decisions for Investors

• Buy CRDO/NVDA on a 3%-5% pullback over the next 2-4 weeks, or express it with call spreads into the next hyperscaler capex updates; upside is meaningful if inference demand keeps compounding, but the trade is invalidated if guidance shows AI spend normalizing faster than expected.
• Relative-value: long P vs short DELL for a 3-6 month AI-infrastructure basket trade. P is the cleaner secondary beneficiary if storage demand proves sticky; the pair breaks if P's mix shifts back toward lower-margin hardware or DELL's AI backlog reaccelerates enough to offset margin pressure.
• Trim or fade PSX into strength with a 1-2 month put spread rather than adding new longs here. Best entry is after another headline-driven spike in crack-spread optimism; the thesis is falsified by a renewed Middle East escalation, but a de-escalation is the more likely medium-term compression catalyst.
• Do not chase EWZ today; put it on a 1-3 month alert list and only add if commodity indicators and power-demand data confirm the Brazil-as-AI-infrastructure narrative. Falsifier: a rolling over of iron ore/copper or a stronger dollar that overwhelms the country beta.